Reinforcing neuron extraction and spike inference in calcium imaging using deep self-supervised denoising

Area of Science:

• Neuroscience
• Biotechnology
• Artificial Intelligence

Background:

• Calcium imaging enables single-cell resolution monitoring of neural circuits.
• Detection noise significantly impacts calcium imaging, especially at high frame rates or low excitation.
• Existing methods often require high signal-to-noise ratio (SNR) data for effective noise reduction.

Purpose of the Study:

• To develop a noise-reduction method for calcium imaging that does not require high SNR observations.
• To improve the accuracy of neuron extraction and spike inference from noisy calcium imaging data.
• To facilitate more robust functional analysis of neural circuits.

Main Methods:

• Developed DeepCAD, a self-supervised deep-learning algorithm.
• Applied DeepCAD for spatiotemporal enhancement of calcium imaging data.
• Evaluated DeepCAD's performance in suppressing detection noise and improving SNR.

Main Results:

• DeepCAD significantly suppresses detection noise in calcium imaging data.
• The method improves the signal-to-noise ratio (SNR) by more than tenfold.
• Enhanced data quality led to improved accuracy in neuron extraction and spike inference.

Conclusions:

• DeepCAD offers an effective solution for enhancing noisy calcium imaging data.
• The method empowers more reliable functional analysis of neural circuits.
• Self-supervised deep learning provides a powerful approach for improving neuroimaging techniques.